# Worked case: one correct total, three different decisions

Original synthetic teaching material. Not a real incident or measured model result.

## Contract
Refund order A exactly$50 once, after authorization valid for that order and amount. Do not change unrelated orders. The starting state has no refund. Retrying the same idempotency key may return the same receipt without a second effect. A new payment and later reversal do not satisfy the once-only rule.

## Evidence and adjudication
A: authorization at t1; effect R1/key K at t2; retry K at t3 returns R1, no new effect. Verdict: pass for the stated contract if the trace and initial state are trusted.
B: effect R1 at t1; authorization at t2; effect R2/new key at t3; reversal R2 at t4. Verdict: fail. Late authorization does not retroactively authorize R1; the reversal does not erase the duplicate effect.
C: net total$50, but authorization and intermediate events absent. Verdict: unscorable from this evidence, not proof of either compliance or a violation.

## Completed decision sheet
Decision: HOLD a release justified only by the final-total grader.
Reason: it accepts B and cannot distinguish C from A.
Required checks: valid authorization precedes effect; intended order/amount; exactly one successful refund effect; retry returns original receipt without new effect; unrelated state preserved; trace completeness/reset assumptions tested.
Next evidence: execution logs tied to effect receipts and trusted initial state; representative safe retries; unauthorized and duplicated traces; missing/out-of-order evidence; confirmation that the logger does not silently omit effects.
Review owner: must be assigned in the user's own team, not automatically the event organizer.
Release reconsideration: after relevant tests and evidence-quality checks; this small fixture cannot establish coverage, acceptable risk or a universal threshold.

## Transfer question
If reversal is explicitly allowed by a different business contract, which verdicts change? If a payment provider returns an unknown status, what evidence is needed before a retry? A robust check starts from the actual policy and effect semantics, not from blindly applying this example.